OpenAI’s latest model, ChatGPT o3, stunned testers by scoring 327 out of 360 on the JEE Advanced 2025 mock test—equivalent to an All India Rank (AIR) 4—a feat revealed by an IIT Kharagpur engineer on June 8, 2025
🧠 Testing Conditions & Methodology
- The test was conducted by IIT‑Kharagpur alumnus Anushka Aashvi, who meticulously recreated real exam conditions: each question was presented in a fresh chat session, with strict instructions—no internet, no Python tools, and no prompts—ensuring an unbiased performance
- Anushka first ran the JEE Advanced paper (held May 18, 2025), prompting the AI to “act like a JEE aspirant.” The model achieved a near-perfect showcase: perfect 60s in both Chemistry & Math (Phase 2) and a near-equal split of 52/49/52 in Phase 1
✅ Performance Highlights
- Score Breakdown: 327/360 marks.
- AIR 4: Equivalent to ranking 4th among over 1.5 million test takers
- Strengths: Flawless problem-solving in lengthy algebra, multi-concept calculus, and visualizing skeletal chemical structures
⚠️ Limitations Noted
- AI stumbled in graphical interpretation and using real-world instruments like the Vernier scale.
- It took nearly ten minutes on a visuals-based question—ultimately incorrect—highlighting current gaps in spatial and tactile reasoning gulte.com
🌐 Implications for AI & Education
- AI leap: Just a few years ago, GPT‑4 would have placed in the 10,000s on the same test—today’s o3 model ranks among the very top
- Human vs AI: This milestone evokes comparisons to IBM’s Deep Blue beating Garry Kasparov—now, AI is excelling not just in chess but academic exams.
- Education disruption: The capabilities of AI in solving complex exam questions raise pressing questions about standardized testing and the future of assessments.
📌 Summary & Outlook
| Aspect | Details |
|---|---|
| Max Score | 360 marks |
| ChatGPT o3 Score | 327 marks |
| Equivalent AIR | 4 (out of 1.5 million+ candidates) |
| Strengths | Algebra, calculus, organic chemistry, multi-concept reasoning |
| Limitations | Graphical visualization, instrument-based questions |
| Broader Impact | Marks a major leap in AI problem-solving, prompting questions on the future of education |
